The Agent Skills Directory

[PROMPT_INJECTION]: A static detector identified an override pattern, but this is a false positive. The skill instructions contain examples of prompt injection (e.g., "Ignore previous instructions") specifically to instruct the agent to refuse such attempts and uphold safety constraints.
[COMMAND_EXECUTION]: The skill executes the official Alibaba Cloud CLI (aliyun) and bundled Python scripts (event_log_analyzer.py, search_filter_generator.py) for diagnostic purposes. These operations are restricted to read-only actions (List, Get) and include strict format validation for identifiers to prevent command injection.
[DATA_EXFILTRATION]: The skill is designed to read Spark application logs and metrics from Alibaba Cloud OSS. It explicitly prohibits the agent from requesting, writing, or handling hardcoded credentials (AK/SK), instead relying on the user's pre-configured CLI profile.
[EXTERNAL_DOWNLOADS]: All external references point to official Alibaba Cloud documentation or trusted binary installation paths for the aliyun CLI. The instructions explicitly forbid downloading or executing external scripts from the internet at runtime.
[REMOTE_CODE_EXECUTION]: The skill specifically prohibits the execution of external code and only permits running the two Python scripts included in its own scripts/ directory. These scripts use only the Python standard library and perform data parsing without dangerous functions like eval() or exec().

alibabacloud-analyticdb-spark-application-analysis-helper